Re: [PATCH v4] thunderbolt: Fix tb->lock deadlock during hot-unplug on AMD USB4 routers

[Mika Westerberg <mika.westerberg@linux.intel.com>]

Hi,

On Fri, Aug 28, 2026 at 12:19:03AM -0500, Juan Martinez wrote:
> Commit f1de1fc5f632 ("thunderbolt: Add quirk to reset host interface on
> DMA path teardown for AMD USB4 routers") introduced a deadlock when
> physically unplugging a Thunderbolt cable on AMD systems.
> 
> The problem occurs because tb_handle_hotplug() holds tb->lock while
> processing the unplug event. When it removes the XDomain services,
> tbnet_remove() calls tb_xdomain_disable_paths() which eventually calls
> tb_domain_reset_interface(). That function tries to acquire tb->lock
> via guard(mutex), but the hotplug worker already holds it, causing a
> self-deadlock.
> 
> The deadlock manifests as a complete network hang because
> tb_handle_hotplug() holds RTNL while waiting on its own mutex, blocking
> all network operations system-wide.
> 
> The existing code already handles this scenario partially: when
> xd->is_unplugged is true, tb_disconnect_xdomain_paths() intentionally
> skips the DMA teardown because the hotplug handler will do it later
> via __tb_disconnect_xdomain_paths(). However, the reset was still
> being called unconditionally.
> 
> Fix this by:
> 1. Splitting tb_domain_reset_interface() into a locked inner function
>    __tb_domain_reset_interface_locked() and a locking wrapper
> 2. Skipping the reset in tb_domain_disconnect_xdomain_paths() when
>    xd->is_unplugged is true (matching the existing teardown skip logic)
> 3. Calling __tb_domain_reset_interface_locked() from tb_handle_hotplug()
>    after __tb_disconnect_xdomain_paths() where the actual DMA teardown
>    happens and tb->lock is already held
> 
> This preserves the reset behavior for normal shutdown paths while
> avoiding the deadlock during physical cable unplug.
> 
> Fixes: f1de1fc5f632 ("thunderbolt: Add quirk to reset host interface on DMA path teardown for AMD USB4 routers")
> Signed-off-by: Juan Martinez <juan.martinez@amd.com>
> ---
>  drivers/thunderbolt/domain.c | 20 +++++++++++++++++---
>  drivers/thunderbolt/tb.c     |  1 +
>  drivers/thunderbolt/tb.h     |  1 +
>  3 files changed, 19 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/thunderbolt/domain.c b/drivers/thunderbolt/domain.c
> index 12c88509a54f..4cef9f4de523 100644
> --- a/drivers/thunderbolt/domain.c
> +++ b/drivers/thunderbolt/domain.c
> @@ -788,14 +788,22 @@ int tb_domain_approve_xdomain_paths(struct tb *tb, struct tb_xdomain *xd,
>  			transmit_ring, receive_path, receive_ring);
>  }
>  
> -static void tb_domain_reset_interface(struct tb *tb)
> +/*
> + * __tb_domain_reset_interface_locked - Reset host interface (lock held)
> + *
> + * Caller must hold tb->lock. Used by hotplug path where lock is already held.
> + */
> +void __tb_domain_reset_interface_locked(struct tb *tb)
>  {
>  	struct tb_nhi *nhi = tb->nhi;
>  
> +	lockdep_assert_held(&tb->lock);
> +
>  	if (!nhi->ops->reset_interface)
>  		return;
>  
> -	guard(mutex)(&tb->lock);
> +	if (!(nhi->quirks & QUIRK_RESET_DMA_ON_TEARDOWN))
> +		return;
>  
>  	/* The reset clears the ring state so stop the control channel */
>  	tb_ctl_stop(tb->ctl);
> @@ -803,6 +811,12 @@ static void tb_domain_reset_interface(struct tb *tb)
>  	tb_ctl_start(tb->ctl);
>  }
>  
> +static void tb_domain_reset_interface(struct tb *tb)
> +{
> +	guard(mutex)(&tb->lock);
> +	__tb_domain_reset_interface_locked(tb);
> +}
> +
>  /**
>   * tb_domain_disconnect_xdomain_paths() - Disable DMA paths for XDomain
>   * @tb: Domain disabling the DMA paths
> @@ -835,7 +849,7 @@ int tb_domain_disconnect_xdomain_paths(struct tb *tb, struct tb_xdomain *xd,
>  	if (ret)
>  		return ret;
>  
> -	if (tb->nhi->quirks & QUIRK_RESET_DMA_ON_TEARDOWN)
> +	if (!xd->is_unplugged)

If I read this right, if you unplug a tree (say a router and then after
that router there is inter-domain link) with this check the reset does not
happen and AMD system still hangs?

>  		tb_domain_reset_interface(tb);
>  
>  	return 0;
> diff --git a/drivers/thunderbolt/tb.c b/drivers/thunderbolt/tb.c
> index b7cc6894a598..89dfb3381345 100644
> --- a/drivers/thunderbolt/tb.c
> +++ b/drivers/thunderbolt/tb.c
> @@ -2489,6 +2489,7 @@ static void tb_handle_hotplug(struct work_struct *work)
>  			tb_xdomain_remove(xd);
>  			port->xdomain = NULL;
>  			__tb_disconnect_xdomain_paths(tb, xd, -1, -1, -1, -1);
> +			__tb_domain_reset_interface_locked(tb);
>  			tb_xdomain_put(xd);
>  			tb_port_unconfigure_xdomain(port);
>  		} else if (tb_port_is_dpout(port) || tb_port_is_dpin(port)) {
> diff --git a/drivers/thunderbolt/tb.h b/drivers/thunderbolt/tb.h
> index 4373336d9425..2e6e0920cb1f 100644
> --- a/drivers/thunderbolt/tb.h
> +++ b/drivers/thunderbolt/tb.h
> @@ -789,6 +789,7 @@ int tb_domain_disconnect_pcie_paths(struct tb *tb);
>  int tb_domain_approve_xdomain_paths(struct tb *tb, struct tb_xdomain *xd,
>  				    int transmit_path, int transmit_ring,
>  				    int receive_path, int receive_ring);
> +void __tb_domain_reset_interface_locked(struct tb *tb);
>  int tb_domain_disconnect_xdomain_paths(struct tb *tb, struct tb_xdomain *xd,
>  				       int transmit_path, int transmit_ring,
>  				       int receive_path, int receive_ring);
> -- 
> 2.43.0